We have used ChIP chip to map, to 1–2-kb resolution, the genome-wide pattern of cohesin association under several different growth conditions (metaphase arrest by cdc16
-ts or nocodazole, galactose versus glucose as a carbon source, and induction of meiosis) and in three different yeast-strain backgrounds (W303a, SK1, and A364a). Using PeakFinder, a program that assesses cohesin binding by comparison of signal to variable local background, we find that the majority of cohesin-binding sites are remarkably constant under these different circumstances. Distribution of cohesins throughout the genome appears to depend on a combination of base composition, sequence, and transcriptional activity. We find evidence for three types of cohesin sites in the genome: (1) the centromere and pericentric domain, (2) intergenic regions in chromosome arms, and (3) ORFs in chromosome arms. The association of cohesin with these three types of sites is subject to different genomic parameters. Cohesin at centromeres and pericentric regions is spread over a broad domain with an elevated “baseline” level and is not affected by the natural transcriptional and coding status. Much of the cohesin in chromosome arms is located in transcriptionally converging intergenic regions. ORFs in chromosome arms where cohesin is found are enriched for repetitive sequences. This suggests that there may be three mechanisms to load cohesin, consistent with what has been proposed for cohesin in meiotic chromosomes for S. pombe
(Kitajima et al. 2003
). A unifying feature of all three types of sites is high AT content.
Pericentric regions contain the most intense and broadest levels of cohesin in the genome (for a more complete analysis of pericentric cohesin see Weber et al. 
). This finding is consistent with a model in which a centromere contains determinants of two opposing processes: (1) pulling the chromosomes apart, via the assembled kinetochore attached to a microtubule, and (2) keeping chromosomes together, via pericentric cohesion. The intensity and breadth of cohesin binding at pericentric regions is similar for all chromosomes, implying microtubules pull all chromosomes with comparable force, regardless of their length. On the other hand, the number of binding sites per chromosome is proportional to chromosome length. This result implies that arm cohesion is not a direct measure of the force exerted by spindle microtubules, and may serve a different function, for instance, to achieve similar levels of condensation. The model in budding yeast that cohesin can participate in genome maintenance in two ways, namely condensation via arm cohesin and biorientation via pericentric cohesin, is intriguing in light of the recent finding that cohesin complexes with different subunits are found on arms and pericentric regions on meiotic chromosomes in S. pombe
and apparently serve different functions (Kitajima et al. 2003
Cohesin cannot stay bound to DNA in the face of active transcript elongation based on three independent cohesin sites (promoter of GAL2,
CARC1, and CARL2). If cohesin and transcript elongation were incompatible, then we would also expect to find sites biased towards intergenic regions, which we do. However, we find a strong bias towards intergenic regions with converging transcription, and a bias against intergenic regions with surrounding unidirectional transcription or diverging transcription, suggesting that intergenic regions with converging transcription may have especially low transcription. These regions may have evolved particularly strong transcriptional stops since they are quite short on average and the cell may need to avoid transcription from one side extending to the other to prevent the synthesis of antisense RNA. The protection of sequence elements important for the replication and segregation of eukaryotic chromosomes from transcription may be a general necessity for their proper function in vivo. For instance, transcription through an autonomous replicating sequence (Snyder et al. 1988
) or a centromere (Hill and Bloom 1987
) disrupts their function.
The observed antagonistic relationship between transcription and cohesin binding in chromosome arms can be explained in two ways. Firstly, transcript elongation may be directly responsible for displacing cohesin. In this type of model, cohesin loading/binding is random, and transcription (and possibly other DNA metabolic processes) “pushes” cohesin into place or strips cohesin from inappropriate locations in each cell cycle. Secondly, transcript elongation may be indirectly responsible for localizing cohesins, for example by accumulation of “nonpermissive” chromatin in transcribed regions and “permissive” chromatin in nontranscribed regions. This type of genome-wide demarcation of transcription units has been shown to occur in S. cerevisiae
(Nagy et al. 2003
) and may depend on nucleosomes (Lee 2004
) and histone variants. The chromatin remodeling complex RSC (Remodels the Structure of Chromatin) has recently been shown to be important for establishment of cohesin in chromosome arms (Baetz et al. 2004
; Huang et al. 2004
). The preferential location of cohesin in heterochromatin in S. pombe
also supports the idea of chromatin modification/structure as the basis for cohesin localization (Bernard et al. 2001
; Nonaka et al. 2002
). The possibility also exists that cohesin itself may influence transcriptional status and act as a transcriptional boundary (Hagstrom and Meyer 2003
; Rollins et al. 1999
Despite the subunit difference between the meiotic and mitotic cohesin complex, we find that the association of cohesin with DNA in meiotic cells is similar to that in mitotic cells. In addition, we find that the constitutive peaks of meiotic cohesin binding are negatively correlated with DSB sites. This negative correlation is consistent with the model proposed by Blat et al. (2002)
for the relationship between recombination and cohesin. This model suggests that cohesins are at meiotic chromosome cores and that recombination occurs in chromatin loops emanating from these cores where the part of the loop undergoing recombination is transiently localized to the axis. Thus, the recombination machinery can sense chromosome organization provided by cohesin. The differences in binding of meiotic and mitotic cohesin suggest that the location of the meiotic complex is also dependent on gene transcription. Hence, meiotic recombination in a given organism may be somewhat dependent on the spacing of cohesin as established in premeiotic S phase, which is in turn responsive to transcription. The genome-wide distribution of DSBs is positively correlated with regions of high GC content, divergent promoters, and transcription factor binding (Gerton et al. 2000
). Thus transcription, recombination, and cohesion all display intimate connections to chromosome and chromatin structure.
Genome-wide studies of protein–DNA complexes afford a better understanding of the role of these complexes in the biology of an organism and its genome. In the process of analyzing the first genome-wide map of cohesin in any organism, we developed PeakFinder, a program able to sensitively identify binding sites of protein–DNA complexes in their local genomic environment, and potentially useful for analysis of any other genome-wide measurements. While budding yeast appears to have largely opted for placement of cohesin in AT-rich, transcriptionally inactive regions, other organisms with much longer and more complicated transcriptional units, different base composition properties, or different levels of condensation may employ different strategies for the placement of cohesin, which may in turn affect the stability of those genomes. The genome-wide analysis of cohesin in S. cerevisiae will serve as a useful framework upon which to explore attributes of cohesin localization in higher eukaryotes.